XP clients sometimes ignore DHCPOFFERs

[John Wobus]

Hi Tom,

I'm cc'ing this to the DHCP list in case others could use the info.
I posted some of this before, but when I was hunting for
solutions online, I came up with very little, it it would be
very good to have a bit about this in the mailing list archive.

We made the change (xx.xx.xx.255 broadcast addresses
to 255.255.255.255 broadcast addresses in the routers)
a few months ago and it appears to have fixed the problem
of Windows clients sometimes ignoring DHCPOFFERS.
We made the change campus-wide, the change forced very little other
reconfiguration: throughout the campus, with tens of
thousands of devices, I heard about just one department
that had to reconfigure just one device that listened
to our routers' RIP broadcast of the default route,
and needed a configuration change to match.

I was loaned a laptop that exhibited the DHCP problem,
thus I had the chance to watch instances of the failure,
try things, do various status commands, capture packets,
etc., and discover what was going on.  When we
reconfigured the broadcast address on one subnet, the
laptop no longer had problems on that subnet.

Then we announced the change to technical staff
across campus and rolled the change out campus wide, but
unfortunately, I receive limited feedback from the campus
at large.  In general, if someone has problems, it might be
reported; if things go right, we're likely to hear nothing.
If we were truly successful eliminating a problem,
we will receive fewer problem reports about our DHCP
service in the future than we would have without
this change.  I asked a few folks who had reported
past problems, but received no definitive feedback regarding
whether the specific problems had gone away.

Someone on staff here recalled that specific past versions of
Windows clients had this particular issue and that
current versions do not.  I did not experiment with
a variety of versions: only one, which I don't recall,
but I'm pretty sure it was not current.  Naturally, given the
solution was so painless that we were going to do it
in any case, spending effort getting the exact list
of clients to be "helped" was of less use.

The failure was non-deterministic: it looked to me like
Windows' DHCP handling raced with its APIPA handling,
and if APIPA won, Windows had an address & mask that
would not accept the router's broadcast address.  The
laptop I borrowed had a tendency to fail but occasionally succeeded.
I can easily believe that the exact timing that determines
whether the failure occurs in any particular instance could
depend upon processor speed, other startup software installed, etc.
I could also easily believe that various Windows versions
have the problem to varying degrees.  However this
is my own speculation.

John

P.S. There are excellent explanations of Windows' APIPA
feature that Googling "APIPA" will turn up.

On Mar 21, 2006, at 9:36 AM, Tom Greaser wrote:

> your quite today.. ive been following this thread.. just wanted to
> know if the bcast fixed your problem.. Im having what seems to be
> the same kinda problem..